Calculating Upper-CL/Lower-CL and Process Capability

D

dschneider-gamra

I am trying to calculate UCL/LCL and process capability and I dont know what data is "correct" to use.

I generate an SPC data charts that are each comprised of thousands of data points. The SPC chart presents range, cp, cpk, max, min, min, std dev.

If I am trying to determine the overall process capability and calculate an UCL/LCL should I use all l of the data points (numbering in the millions of data points) and generate the the statistical data from all of that data. (this would be very labor intensive to get all the data into one spreadsheet).

Or

Do I use the max/min range from the SPC data?

or

Do I use the Mean +/- 3 std dev

or

:magic: do I perform a majic spell on the data....

Those are all the ideas that I have - hopefully I have supplied enough information on the data that I have and the question. If not please let me know.

Thanks for the help.

Dave
 

Marc

Fully vaccinated are you?
Leader
A quick "Bump". My Thanks in advance to any of you can help with this one.
 
A

AdamP

I generate an SPC data charts that are each comprised of thousands of data points. The SPC chart presents range, cp, cpk, max, min, min, std dev.

Well if I read your post correctly, you have some of your answer already. If your chart provides both Cp adn Cpk, assuming your data is normally distributed, you have capability reported. Do go ahed and check the normality though.

Can you tell us what software is used for the 'SPC' chart? With everything else it reports, it seems like there might also be control limits somewhere. If not, then you'll want to look at the data sampling structure to determine the most correct way to determine the control limits.

Cheers,

Adam
 

Jim Wynne

Leader
Admin
I am trying to calculate UCL/LCL and process capability and I dont know what data is "correct" to use.

I generate an SPC data charts that are each comprised of thousands of data points. The SPC chart presents range, cp, cpk, max, min, min, std dev.

If I am trying to determine the overall process capability and calculate an UCL/LCL should I use all l of the data points (numbering in the millions of data points) and generate the the statistical data from all of that data. (this would be very labor intensive to get all the data into one spreadsheet).

Or

Do I use the max/min range from the SPC data?

or

Do I use the Mean +/- 3 std dev

or

:magic: do I perform a majic spell on the data....

Those are all the ideas that I have - hopefully I have supplied enough information on the data that I have and the question. If not please let me know.

Thanks for the help.

Dave

I'm not sure where to begin but, if you can answer some questions you'll get better help:
  1. What's the manufacturing process?
  2. How long does it take to accumulate millions of data points?
  3. Are you sure you're dealing with a normal distribution?
  4. You say you've generated an SPC chart. Have you used rational subgrouping or individuals? In either case, were samples taken at uniform intervals and charted in chronological order?
That's enough for now, I think.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I'm not sure where to begin but, if you can answer some questions you'll get better help:
  1. What's the manufacturing process?
  2. Are you sure you're dealing with a normal distribution?
  3. You say you've generated an SPC chart. Have you used rational subgrouping or individuals? In either case, were samples taken at uniform intervals and charted in chronological order?
  4. Can you supply some sample data?

This should be a boilerplate for all SPC questions! :tg:
 
D

Darius

:magic: do I perform a majic spell on the data....

I like that:lmao:

Normally, I would try the data it self, altho you mention Ranges and Min/Max as I can understand the data is charted as IXmR.

As Donald Wheeler said (more or less), there are not right and wrong Control Limits, the control limits are a reflexion of the process natural variation .

You can take "n" data points (let's say 120 data points) and obtain the statistics, and take another set of the data and compare the results, if the process looks more or less the same, you can take them as a good estimate.

The experts say that you shouldn't recalculate the control limit's unless there is a change on the behavior of the data (aka change of the mean, trends etc).
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I also recommend looking at this, realizing that collecting the data is in the middle of the SPC implementation process, not the beginning.
 
D

dschneider-gamra

1) the mfg process is profile extrusion of plastic.
2) a data point is taken every 6" and we run 60'/min - so we get 120 data points each minute.
3) Each data point is ploted and the data is generated - max/min/cp/cpk etc.
4) not sure how to input data - I tried to insert but the post was rejected:mad:

I know I get a CP/CPK for each run - how do I determine overall capability? How to I set an overall UCL/LCL based on all the data points - or do I use some of the calculated values.

The customer is saying we should have control limits +/-0.010" - I am trying to show them statistically that the control limits need to be calculated based on the data generated not a number pulled out of a hat.

By the way I should mention that my personal cabability (pc not cp) on understanding statistics = 0.002 :eek:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I know I get a CP/CPK for each run - how do I determine overall capability? How to I set an overall UCL/LCL based on all the data points - or do I use some of the calculated values.

If you can use all of the data (not limited by the number of data points), then I would use them all.

The customer is saying we should have control limits +/-0.010" - I am trying to show them statistically that the control limits need to be calculated based on the data generated not a number pulled out of a hat.

Yeah, what is a customer thinking "specifying" control limits? That is an abject lack of SPC knowledge...unless they did some analysis of the data you gave them at some point in time.
 
Top Bottom